Kelly Criterion for Sports Betting — Size Your Edge, Not Your Ego

Most bettors have a staking problem that isn't the obvious one. They don't bet recklessly on every match. They bet inconsistently — big on gut feelings, small when unsure, with no mathematical connection between bet size and perceived edge. The result is that even a bettor with genuine edge can blow up their bankroll through poor sizing before the edge has time to materialise.

The Kelly Criterion solves this. It's a formula for optimal bet sizing — not optimal in the sense of "most fun" or "most comfortable," but optimal in the mathematical sense: maximising the long-run growth rate of your bankroll.

What Problem Kelly Actually Solves

Before Kelly (1956), there was no rigorous answer to: "I have a 55% edge. How much should I bet?"

The intuitive answers are both wrong:

• Bet as much as possible — maximises short-term gain but guarantees eventual ruin (any bad run wipes you out)
• Bet a flat small amount — safe but leaves massive growth on the table

Kelly found the exact staking fraction that maximises the geometric mean of outcomes — the growth rate that matters in reality, where results compound over time. At Kelly, you never risk ruin from a single bet, and you grow faster than any other consistent strategy.

The Formula

f* = (bp - q) / b

Where:

• f* = fraction of bankroll to bet
• b = net odds received (decimal odds minus 1)
• p = your estimated probability of winning
• q = probability of losing = 1 - p

In plain English: Kelly tells you to bet the fraction of your bankroll where your edge (the numerator) is proportional to the reward (denominator). More edge and better odds = larger bet. Less edge or short odds = smaller bet.

Worked Example

You've assessed a match and believe Team A has a 58% chance of winning. The bookmaker is offering 2.10 decimal (after de-vigging, you believe the true price should be 1.72, so there's genuine value here).

b = 2.10 - 1 = 1.10
p = 0.58
q = 1 - 0.58 = 0.42

f* = (1.10 x 0.58 - 0.42) / 1.10
f* = (0.638 - 0.42) / 1.10
f* = 0.218 / 1.10
f* = 0.198 -> ~20% of bankroll

Full Kelly says bet 20% of your bankroll. If you have $1,000, that's a $200 bet.

That probably sounds terrifying. It should. And that's exactly why we need to talk about fractional Kelly.

Why Full Kelly Is Too Aggressive

Kelly is mathematically optimal only if your probability estimate is perfectly accurate. In practice, it never is.

Probability estimation in sports betting involves genuine uncertainty — injury news you might have misweighted, historical data that may not apply to this specific matchup, line movement you haven't fully interpreted. Even skilled bettors carry meaningful error in their probability estimates.

Full Kelly with overestimated edge is a reliable path to large drawdowns. The Kelly curve is not symmetric: overbetting by 10% reduces long-run growth significantly more than underbetting by 10% does. Mistakes on the upside hurt you more than caution.

Additionally, even correct full Kelly produces brutal variance. A sequence of Kelly-sized bets can see 30-50% drawdowns that are mathematically expected — and psychologically devastating. Most people abandon their strategy long before the edge has time to compound.

Fractional Kelly: The Practical Default

The fix is straightforward: bet a fraction of the Kelly recommendation. Common fractions:

• Half Kelly (0.5x): Reduces variance to 25% of full Kelly variance, while retaining about 75% of the growth rate. The most popular practical choice.
• Quarter Kelly (0.25x): Very conservative, smooth equity curve, suitable for bettors still calibrating their edge.

Returning to our example:

Half Kelly at $100 still grows your bankroll faster than flat staking — but the ride is dramatically smoother, and you stay in the game through inevitable losing runs.

Most professional bettors and staking models operate between 0.25x and 0.5x Kelly. Anything above 0.5x requires extremely high confidence in your probability estimates.

Garbage In, Garbage Out

Here's the part most articles gloss over: Kelly is only as good as your probability estimate.

The formula is mechanical. Feed it your edge and it spits out a stake. But if your 58% is actually 52% (a modest overestimate), the "edge" disappears or reverses:

f* = (1.10 x 0.52 - 0.48) / 1.10
f* = (0.572 - 0.48) / 1.10
f* = 0.092 / 1.10
f* = 0.084 -> ~8.4%

The bet still exists, but at less than half the recommended size. And if your estimate is below the break-even probability (where bp = q), Kelly correctly outputs a negative number — meaning don't bet at all.

This is why systematic probability estimation — using a model, tracking your historical accuracy, comparing against closing lines — matters far more than the staking formula. Kelly amplifies skill; it doesn't create it.

The Hard Cap Rule

Regardless of what Kelly outputs, many professional bettors impose a hard cap on any single bet — typically 2-5% of bankroll, sometimes up to 10% for the highest-conviction plays.

A reasonable rule: never bet more than 25% of bankroll on a single event, full stop. This is a protection against model failures and black swan events that probability estimates can't capture. It also prevents the psychological damage of a catastrophic single loss from derailing your strategy.

In practice, with half Kelly staking and reasonable odds, you'll rarely approach this cap unless you're estimating very high probabilities. But the rule exists for the edge cases.

Kelly Is Math, Not Magic

A responsible framing of the Kelly Criterion: it tells you how to size bets once you have genuine edge. It does not manufacture edge. It does not tell you which bets to take. And it does not protect you from the core risk in sports betting — that you're wrong about having edge in the first place.

Use Kelly as a disciplining framework: it forces you to quantify your confidence before committing capital. If you can't articulate your probability for a bet, you shouldn't be placing it. If your estimated probability barely exceeds the break-even threshold, Kelly will correctly suggest a tiny stake — and that friction is valuable.

Track your closing line value. If your bets consistently beat the closing line after de-vigging, that's your signal that estimated probabilities are in the right neighbourhood. If they don't, reduce stakes further while you recalibrate.

The formula has been around since 1956. Gamblers and investors have tested it extensively. The conclusions are consistent: the growth rate advantage is real, the variance without fractional adjustment is brutal, and the estimation problem is the actual hard part. Everything else is arithmetic. Size your next bet with our free Kelly Criterion Calculator.